The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes Git and GitHub CLI (gh) commands to manage PRs and local workspaces. All commands are executed with structured arguments as lists, avoiding shell injection risks. The use of subprocess.run in Python scripts is limited to these trusted tools.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from PR descriptions, review comments, and associated issues. It mitigates injection risks by explicitly instructing the agent to treat this content as evidence rather than instructions (references/problem-review.md). It also uses structured JSON snapshots and regular expressions to validate repository and issue identifiers.
[DYNAMIC_EXECUTION]: The skill implements a self-integrity mechanism in scripts/prepare_helper_fingerprint.py that calculates SHA-256 hashes of its own scripts to ensure consistency across review phases and detect unauthorized modifications.
[DATA_EXFILTRATION]: Network activity is restricted to communication with GitHub (a well-known service) via the authenticated gh CLI. The skill does not access sensitive local files like SSH keys or environment secrets.
[PRIVILEGE_ESCALATION]: Git operations are performed within the local repository or isolated worktrees. The skill requires a clean workspace before switching branches and verifies upstream targets to prevent unauthorized modifications to the local environment.

review-pr